Subcorp is a self-hosted multi-agent AI platform I built to explore how autonomous agents can collaborate, use tools, retain context, and drive work forward inside a full-stack product.
It brings together a modern web application, typed backend services, real-time communication, vector-backed memory, and containerized infrastructure in one cohesive system.
- Full-stack engineering: built with Next.js, React, TypeScript, and PostgreSQL across both product and platform layers
- AI systems design: orchestrated six distinct agents with role-based behavior, memory, tool access, and long-running workflows
- Backend architecture: modeled proposals, missions, events, and execution pipelines for autonomous task handling
- Real-time product development: implemented live multi-agent interaction over WebSockets
- DevOps mindset: packaged the stack with Docker Compose and repeatable build and operational workflows
- Next.js 16
- React 19
- TypeScript
- PostgreSQL 16 + pgvector
- WebSockets
- OpenRouter SDK
- Tailwind CSS 4
- Docker / Docker Compose
- Designed a closed-loop workflow for proposals, missions, roundtables, and agent sessions
- Built native tool execution for web research, file access, shell commands, and delegated sub-agents
- Combined frontend UX, backend orchestration, data modeling, and infrastructure in a single production-oriented codebase
This project is a strong example of the work I enjoy most: turning ambitious product ideas into reliable software, connecting AI capabilities to practical architecture, and shipping across the stack.
If you're hiring for full-stack, platform, or AI-enabled product engineering, Subcorp is meant to show both technical range and the ability to execute on complex systems.
/srccontains the application, APIs, and core product logic/scriptscontains worker and operational tooling/docscontains deeper architecture and project notes